
Face-centered cubic (FCC) lattice models for protein folding: energy function inference and biplane packing Author: Allan Stewart Adviser: Sorin Istrail, Second Reader: Franco P. Preparata 1 Abstract Protein Structure Prediction (PSP) is a grand challenge of bioinformatics with broad implications for com- binatoric optimization, computational geometry, and physical energy models. Under most accepted compu- tational models, PSP is NP-hard, even when the chemical and physical properties of proteins are greatly simplified. The objective of PSP (also known as protein folding) is to select the molecule conformation which minimizes the energetic potential. We study real biological proteins from the Protein Data Bank in order to infer general energy functions for the protein folding problem. While the general problem is intractable, we seek to find methods of folding and mathematical principles that might suggest how to optimize sufficiently realistic energy functions. We will study the established HP model on the face-centered cubic lattice, con- jectured by Kepler and proven by Hales to provide the \tightest" packing of identical spheres. The objective of analyzing Protein Data Bank structures is two-fold. We firstly seek the best possible energy function for this model. Secondly, we will investigate biplanar and octahedral structures in the FCC lattice which may yield novel algorithms for structure prediction. Face-centered cubic (FCC) lattice models for protein folding { Allan Stewart Contents 1 Abstract 0 2 Introduction 2 2.1 Hardness results for protein folding . 4 2.2 Bridging the lattice debate . 5 2.3 Method of Protein Chain Lattice Fitting . 6 3 The Protein Energy Potential Function Problem 8 3.1 Introduction to the energy function inference problem . 8 3.2 Construction of the HP Model . 10 3.3 Model fitting in general . 12 3.4 Model inference for the FCC-fit protein dataset . 15 3.5 Pairwise Function Impossibility Conjecture [2] . 18 4 Modelling Hydrophobic Collapse 24 4.1 The State of the Art . 24 4.2 Evaluating the biplanar model for hydrophobic collapse . 27 4.3 α-helices and polyhedra in the FCC lattice . 31 4.4 Optimal Bipole Packing: folding's paralog . 34 4.5 Discussion and caveats: Algorithms on discrete lattice . 37 5 Acknowledgements 39 6 Bibliography 40 7 Appendices 43 7.1 Proof sketch of approximation ratio for the Stewart FCC-sidechain algorithm . 43 7.2 A protein and its decoy: 8RXN . 46 7.3 PDB IDs Associated with Biplanar and Non-Biplanar Structure . 47 1 2 Introduction Proteins are the primary functional units of the living cell. The challenge of inferring a protein's three- dimensional structure from its sequence is known as the Protein Structure Prediction (PSP) problem. PSP is of crucial importance to the biological community, which has collected protein structures in databases since the 1960s. Chemical methods of measuring protein conformations (such as X-ray crystallography) have improved over time and contributed to the founding of large protein structure databases such as the Protein Data Bank (PDB). Nevertheless, there exist classes of proteins for which 3D structure reconstruction is impossible using these methods. In addition, scientists wish to discover first principles of protein folding (\folding" and \structure prediction" are used interchangeably throughout this paper, although folding refers to a larger body of protein-related problems). Solving the protein folding problem would enhance understanding of diseases which are caused by misfolding and allow scientists to speculate on unknown proteins (ie. drug design). For these reasons, PSP is a long-standing and elusive problem in computational biology. Protein structure is diverse, but by no means arbitrary. Biologists classify proteins as polypeptide polymers which form a backbone chain by the dehydration of their carboxyl group. Each amino acid carries a sidechain functional group, except for the compact amino acid glycine. The primary structure of the protein is the sequence of these amino acid residues from amino terminus to carboxyl terminus. The primary structure is thus a string over the twenty-letter alphabet of amino acid types. The secondary structure of the protein is defined by domains which exhibit certain structural motifs that stabilize the protein's conformation. The most frequent of these are the right-handed α-helix and the β-sheet, a planar motif in which the backbone threads back around itself. The goal of PSP is to predict the tertiary structure of proteins, which is the three-dimensional shape of the folded protein. Of biological relevance, but beyond the scope of this paper, is quaternary structure, in which several polypeptide subunits interact to form a stable conformation together.[3] The dogma of PSP is that biology has selected the lowest possible energy conformation of the polypeptide as native. The problem is thus to infer the lowest energy conformation from the energy landscape. Molecular biology has accumulated a corpus of knowledge on the commonalities and eccentricities of 2 biological proteins (ie. those present in cells). Several secondary structural motifs are known. The afore- mentioned α-helix is a right-handed helical structure spaced by 3.6 residues per turn, and in which the structure requires certain sidechains and hydrogen bonding to be present. The β-sheet consists of multiple strands which are attracted by hydrogen bonds in either a parallel or antiparallel conformation. Less com- mon, but prominent, formations such as the greek-key and β-hairpin are observed. For membrane-spanning proteins (transmembrane proteins) and the TIM barrel, both α and β domains interact. The secondary structure motifs play a significant role in the tertiary structure, for they define the dihedral angles between adjacent amino acid residues. The polypeptide backbone adopts two torsional angles for each residue: φ for the rotation between the α-carbon and the amine nitrogen atom, and between the α-carbon and the adjacent carbon0 (\carbon-prime", in the carbonyl group).[3] Ramachandran showed that the joint distri- bution of these two angles clusters at three disparate domains, each corresponding to its own secondary structural element.[29] Frequently, scientists use this prior knowledge in order to inform models of protein folding. Nevertheless, PSP is a grand problem which is difficult even when secondary structure information is known beforehand. In fact, Levinthal estimated that under restriction of φ and angles to realistic values, a polypeptide of length 100 has over 10143 possible conformations.[20] It is an understatement that na¨ıve enumeration is no solution for protein folding. One approach to PSP, which we focus on, is ab initio protein folding. Ab initio requires that no prior knowledge of the protein is known except for its primary structure. In practice, these efforts have been less fruitful than which exploit a priori knowledge such as the protein type; for example, the globular protein hemoglobin may be better targeted using a special method. Ab initio methods seek to discover general characteristics of protein structure and are the subject of decades of work. A landmark study by Anfinsen in the 1960s established that a protein denatured (unfolded) by interfering agents reconstituted its native confirmation when those agents were removed.[1] This experiment develops the intuition that protein structure is dependent only on the primary structure and justifies the ab initio effort. 3 2.1 Hardness results for protein folding Computationally, it is generally understood that PSP is an NP-hard problem. Arguably, the most general formulation (Ngo and Marks) postulates a protein in which each atom is identified, and their respective connectivities are enumerated. Given this information alone, and a generic energy function similar to all- atom physical models, U: X bond 0 2 X angle 0 2 X torsion 0 X nonlocal 0 U = Kb (lb−lb ) + Ka (θa−θa) + Kt (1−cos nt(φt − φt ) )+ Kij f(rij=rij) (1) b a t i>j Ngo and Marks showed that the minimization of U is NP-hard for any dimensionless function f(x) with a unique minimum at x = 1.[23] The proof of NP-hardness supposes a polypeptide chain in which a variably folded region (one with several possible torsions) exists, and the global minimum is only achievable when some scaffold region comes in contact with the variable region. The scaffold region must bend around a fixed domain in order to make this contact. Ngo and Marks arrive at a reduction to the partition problem: the energy is minimized if and only if the algorithm answers the Partition problem.[23] The Partition problem asks whether a set of integers may be partitioned into two subsets of equal sum, and is necessary, under the conditions described, for the geometric conformation that is energetically minimum. This formulation corresponds to the general chemical and geometric characteristics of the problem and is respected as a proof of hardness; indeed, there is yet to be an popular formulation which is computationally \easy". For instance, discrete models of protein folding are also NP-hard by reduction to bin packing (the minimization of bins into which objects are placed) or Hamiltonian path problems.[16] Furthermore, PSP is a problem for which there is no unique formulation. There are several models of protein folding, and they generally fall into one of two categories: off-lattice and on-lattice. Off-lattice models allow the protein's components to move, free-floating, in a continuous space. These models are popular in biology since they agree with the observed flexibility of proteins and the science of X-ray crystallography. On-lattice models map the protein's components to points on a discrete lattice. The goal of on-lattice models is to confine the size of the energy landscape and reduce the protein folding problem to its simplest 4 formulation. 2.2 Bridging the lattice debate PSP research is sharply divided between the off-lattice and on-lattice factions.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages50 Page
-
File Size-